1. Introduction
The interior of a star or planet can be divided into convective and radiative layers. The convective layer is unstable to convection, and thus, the energy from the interior of star can be efficiently transported by convection. The radiative layer, on the other hand, is stable to convection, and the energy is mainly transported by radiation. Many stars and planets have adjacent convective and radiative layers. For example, intermediate-mass main-sequence stars (mass is within 0.5–1.5 solar masses) have an outer convective layer adjacent to an inner radiative layer, while massive main-sequence stars (mass is above 1.5 solar masses) have an outer radiative layer adjacent to an inner convective layer (Kippenhahn, Weigert & Weiss Reference Kippenhahn, Weigert and Weiss1990). Gas giant planets, such as hot Jupiters, are also believed to have adjacent radiative and convective layers. The convective stability of a convective or radiative layer can be measured by the square of buoyancy frequency
$N^2$. The convective layer has
$N^2\leqslant 0$ (
$N^2=0$ means neutrally unstable), while the radiative layer has
$N^2>0$. As convection can transport energy efficiently, the thermal structure in the convective layer is almost adiabatic, which yields
$N^2\approx 0$. Wave propagation is an important physical process in stars and planets. In the radiative layer, gravity waves can be excited, because the buoyancy force can act as a restoring force of perturbations from the thermal equilibrium state. In contrast, gravity waves cannot survive in the convective layer. When a gravity wave propagates from a radiative layer into its adjacent convective layer, the wave amplitude will decay quickly. In the presence of rotation, inertial waves can be excited because the Coriolis force can act as a restoring force. Unlike gravity waves, inertial waves can survive in both radiative and convective layers. Gravity waves in a rotating system are sometimes called inertia-gravity waves or gravito-inertial waves. In a rotating system, inertial waves can propagate from a convective layer into a radiative layer as gravito-inertial waves, and vice versa. The purpose of this paper is to investigate wave propagations near radiative–convective boundaries in a rotating system.
It is well known that internal gravity waves (IGW) are likely to be generated near the radiative–convective boundary in stars by the shear stress (Kumar, Talon & Zahn Reference Kumar, Talon and Zahn1999) or stochastic overshooting plumes (Rogers & Glatzmaier Reference Rogers and Glatzmaier2005; Mathis, Neiner & Tran Minh Reference Mathis, Neiner and Tran Minh2014). Various numerical simulations have verified that IGW can be excited efficiently near radiative–convective boundaries in turbulent convection (Meakin & Arnett Reference Meakin and Arnett2007; Brun, Miesch & Toomre Reference Brun, Miesch and Toomre2011; Alvan, Brun & Mathis Reference Alvan, Brun and Mathis2014). When IGW propagate in the radiative stable layer, it could induce material mixing (Rogers & McElwaine Reference Rogers and McElwaine2017) and transport angular momentum (Rogers et al. Reference Rogers, Lin, McElwaine and Lau2013; Fuller et al. Reference Fuller, Lecoanet, Cantiello and Brown2014). For a stable layer above a convective layer, gravity waves are important because they propagate upward into a decreasing density region, which yields increasing wave amplitudes and wave breaking (Rogers et al. Reference Rogers, Lin, McElwaine and Lau2013). In the Earth's atmosphere, it has been confirmed that the breaking of gravity waves is one of the major factors for the transport of water from the troposphere into the stratosphere (Qu et al. Reference Qu, Huang, Vaillancourt, Cole, Milbrandt, Yau, Walker and de Grandpré2020). Using the analogy of Earth's atmosphere, Rogers et al. (Reference Rogers, Lin, McElwaine and Lau2013) proposed that IGW are dynamically and chemically important in massive stars. Bowman et al. (Reference Bowman2019) detected low-frequency variability in massive stars by asteroseismology, and they interpreted it as a signal of IGW. Lecoanet et al. (Reference Lecoanet, Cantiello, Quataert, Couston, Burns, Pope, Jermyn, Favier and Le Bars2019) argued that this observed low-frequency variability might be produced by the subsurface convection, rather than IGW excited by core convection. Wave propagation also plays an essential role on angular momentum transport. Zahn, Talon & Matias (Reference Zahn, Talon and Matias1997) developed a theoretical model on the angular momentum transported by waves. For stars, the model of Fuller et al. (Reference Fuller, Lecoanet, Cantiello and Brown2014) predicts that IGW tend to reduce differential rotation (different parts of a star have different angular velocities) on short time scales in low mass stars. Thus it helps to explain the observed small amounts of internal differential rotation in low-mass main-sequence stars. For planets, it has been shown that IGW possibly play important roles in planetary formations and evolutions (Rogers, Lin & Lau Reference Rogers, Lin and Lau2012; Yu Reference Yu2020).
Apart from gravity waves, in rotating stars or planets, inertial waves could also be generated because of the presence of Coriolis force (Ogilvie & Lin Reference Ogilvie and Lin2004; Wu Reference Wu2005a; Goodman & Lackner Reference Goodman and Lackner2009). Gravity waves can only survive in convectively stable layers, while inertial waves can survive in both convectively stable and unstable layers. Wu (Reference Wu2005b) found that resonantly exited inertial modes have significant impact on the tidal dissipation of a coreless Jupiter (Wu Reference Wu2005b). Goodman & Lackner (Reference Goodman and Lackner2009) argued that a rigid core should be included in the model. In the discussion of Goodman & Lackner (Reference Goodman and Lackner2009), inertial waves were assumed to be fully reflected at the surface of the rigid core. However, recent study (Liu et al. Reference Liu, Hori, Müller, Zheng, Helled, Lin and Isella2019) revealed that Jupiter probably has a diluted core formed by a giant impact. If the rigid core is replaced by a diluted core, inertial waves are expected to be partially reflected at the surface. An analogy can be readily drawn between this problem and that of wave propagations near radiative–convective boundaries.
A question that remains unclear is how waves reflect and transmit near radiative–convective boundaries in stars and planets. Wei (Reference Wei2020a,Reference Weib) has discussed the reflection and transmission of an incident wave at the radiative–convective boundary on the f-plane, with the assumption that the buoyancy frequency abruptly changes from zero at the convective layer to a constant value at the radiative stable layer. He discovered that waves can efficiently transmit across the boundary in a rapidly rotating fluid. In real stars or planets, transitions of thermal structures are likely to be continuous. For example, the transition of static stability profile in Venus's atmosphere is abrupt at middle and polar latitudes, but smooth at equatorial latitudes (Tellmann et al. Reference Tellmann, Pätzold, Häusler, Bird and Tyler2009). Borovikov (Reference Borovikov1990) has considered the trajectory rays of internal waves in a monotonically varying
$N^2$ profile. Lecoanet & Quataert (Reference Lecoanet and Quataert2013) investigated IGW excited by turbulent convection in discontinuous and smooth transitions near radiative–convective boundaries. They found that the flux of IGW in the smooth transition case is larger than that calculated in the discontinuous transition case. Therefore, it is necessary to consider the wave reflection and transmission near a continuously transited radiative–convective boundary. In this paper, we extend the work of Wei (Reference Wei2020a), by considering two different stratification structures in the stable layer: a uniform stratified layer, or a continuously varying stratified layer. In particular, we consider wave transmissions near radiative–convective boundaries under the non-traditional f-plane approximation. In the non-traditional f-plane, both horizontal and vertical Coriolis parameters are set to constant values at a given latitude. It degenerates to a traditional approximation when the horizontal Coriolis parameter is set to zero. The method is described in § 2. The results are reported in § 3. Efficiencies of wave transmissions are estimated and compared in §§ 3.1–3.3. Differences on wave transmissions with and without traditional approximations are investigated in § 3.4. The conclusions and discussion are given in § 4.
2. Method
For a Boussinesq flow on a rotating f-plane, the linearised hydrodynamic equations of mass and momentum conservations (Phillips Reference Phillips1966) are
where
$\boldsymbol {u}=(u,v,w)$ is the velocity;
$\boldsymbol {f}=(0,\tilde {f},f)$ is a vector form of horizontal and vertical Coriolis parameters, with
$\tilde {f}=2\varOmega \sin \theta$ and
$f=2\varOmega \cos \theta$ (
$\,f$ and
$\tilde {f}$ are constants at a given
$\theta$);
$\varOmega$ is the rotation rate and
$\theta$ is the colatitude of the f-plane;
$p$ is the modified pressure perturbation (pressure perturbation scaled by constant background density);
$b$ is the buoyancy;
$N^2$ is the square of buoyancy frequency;
$\hat {\boldsymbol {x}}$,
$\hat {\boldsymbol {y}}$ and
$\hat {\boldsymbol {z}}$ are the unit vectors in the west–east direction, the south–north direction, and the vertical direction, respectively. We consider a two-layer thermal structure, with an upper density-stratified stable layer adjacent to a lower adiabatically convective layer (figure 1a). With this setting,
$N^2$ is positive in the stable layer and zero in the convective layer, respectively.
Sketch plots of f-plane structure and wave direction. (a) Illustration of a two-layer setting f-plane. The stable layer sits above the convective layer. The f-plane is inclined an angle
$\theta$ with the north pole.
$\hat {\boldsymbol {y}}$ is the south-north direction. Here
$\hat {\boldsymbol {z}}$ is the vertical direction,
$\hat {\boldsymbol {x}}$ is the west–east direction pointing into the plane of the paper. (b) The plane wave propagates in a direction at an angle
$\alpha$ to the
$x$-axis. (c,d) The sign of
$c_{p_{0}}c_{g}$ in weakly (case I) and strongly (case II) stratified cases, respectively. Note that
$f^2$ can be smaller than
$N^2$ in the weakly stratified case.

2.1. Wave dispersion relation and frequency range
After some manipulations, the above equations can be reduced into the following equation on vertical velocity (Gerkema & Shrira Reference Gerkema and Shrira2005):
where
$\boldsymbol {\nabla }_{h}^2=\partial _{x}^2+\partial _{y}^2$ is the horizontal Laplacian operator and
$\nabla ^2=\boldsymbol {\nabla }_{h}^2+\partial _{z}^2$ is the Laplacian operator. For a plane wave propagating in the direction
$(\cos \alpha ,\sin \alpha )$, the problem can be further simplified by introducing a new variable
$\chi$, which satisfies
$x=\chi \cos \alpha$ and
$y=\chi \sin \alpha$ (figure 1b). Substituting
$\partial _{x}=\cos \alpha \partial _{\chi }$,
$\partial _{y}=\sin \alpha \partial _{\chi }$ and
$w=W(\chi ,z)\exp (-\textrm {i}\sigma t)$ into the above equation, we obtain
where
$\sigma$ is the time frequency,
$A=\tilde {f}_{s}^2-\sigma ^2+N^2$,
$B=f\tilde {f}_{s}$,
$C=f^2-\sigma ^2$ and
$\tilde {f}_{s}=\tilde {f}\sin \alpha$. A wave solution exists when this partial differential equation is hyperbolic, thus it requires
$\varDelta =B^2-AC>0$ (Gerkema & Shrira Reference Gerkema and Shrira2005), or equivalently
Then the frequency range for wave solutions is
where
$\sigma _{1,2}^2=[(f^2+\tilde {f}_{s}^2+N^2)\mp \sqrt {(f^2+\tilde {f}_{s}^2+N^2)^2-4N^2 f^2}]/2$. For a wave to propagate through the whole domain, this inequality should be satisfied for all the possible values of
$N^2(z)$. Let
$N_{min}^2$ and
$N_{max}^2$ be the minimum and maximum values of
$N^2(z)$, respectively. For a wave to survive in both convective and stable layers, it requires
where
$\sigma _{min}^2=[(f^2+\tilde {f}_{s}^2+N_{max}^2)- \sqrt {(f^2+\tilde {f}_{s}^2+N_{max}^2)^2-4N_{max}^2 f^2}]/2$ and
$\sigma _{max}^2=[(f^2+\tilde {f}_{s}^2+N_{min}^2)+ \sqrt {(f^2+\tilde {f}_{s}^2+N_{min}^2)^2-4N_{min}^2 f^2}]/2$. In our problem
$N_{min}^2=0$, thus
$\sigma _{max}^2=f^2+\tilde {f}_{s}^2$. The width of the frequency range is
This width is always positive unless
$\tilde {f}_{s}^2=0$ and
$f^2+\tilde {f}_{s}^2\leqslant N_{max}^2$. The first condition
$\tilde {f}_{s}^2=0$ can be achieved only when
$\sin \theta =0$ or
$\sin \alpha =0$. Thus, in polar regions, waves cannot survive in both convective and stable layers in a slowly rotating fluid (
$N^2/4\varOmega ^2 \gg 1$, or in other words, the stable layer is strongly stratified) (Wei Reference Wei2020a). However, the conclusion at other latitudes is completely different.
At other latitudes, if the meridional wavenumber is non-zero, waves can survive in both convective and stable layers even in a slowly rotating fluid. From (2.9), we see that the width of frequency range varies with the degree of stratification
$N_{max}^2/(4\varOmega ^2)$, the colatitude
$\theta$ and the plane wave propagating angle
$\alpha$. The dependence of the width of frequency range on these parameters can be studied by analysing the monotonicity of (2.9). After some trivial calculations (see Appendix A) we have the following conclusions. First, the width of the frequency range decreases with
$N_{max}^2/(4\varOmega ^2)$, with an upper limit value
$f^2+\tilde {f}_{s}^2$ achieved at
$N_{max}^2/(4\varOmega ^2)\rightarrow 0$ and a lower limit value
$\tilde {f}_{s}^2$ achieved at
$N_{max}^2/(4\varOmega ^2)\rightarrow \infty$. As a result, waves are more likely to survive at low latitudes in a slowly rotating fluid, where
$\tilde {f}_{s}^2$ is larger. In a rapidly rotating fluid (
$N^2/4\varOmega ^2 \ll 1$, or in other words, the stable layer is weakly stratified), waves are more likely to survive at high latitudes, but the chance at low latitudes is still moderate if the meridional wavenumber dominates the zonal wavenumber. This conclusion shows agreement with the result obtained in the spherical geometry by Friedlander & Siegmann (Reference Friedlander and Siegmann1982) (see their figures 1 and 3). Second, the width of frequency range increases with
$\sin ^2\alpha$, which indicates that waves with higher meridional wavenumbers are more likely to survive. Third, the variation of the width of frequency range on
$\theta$ is not always monotonic. It depends on the degree of stratification. If
$N_{max}^2/4\varOmega ^2>1$, then the width increases with
$\theta$. If
$N_{max}^2/4\varOmega ^2 < 1$, however, there exists a critical value
$(\sin ^2\alpha )_{c}$, and the width increases with
$\theta$ in the range
$\sin ^2\alpha >(\sin ^2\alpha )_{c}$, and decreases with
$\theta$ in the range
$\sin ^2\alpha <(\sin ^2\alpha )_{c}$. Again, we have verified that the frequency range is wider at low latitudes in a slowly rotating fluid. In a rapidly rotating fluid, on the other hand, waves can survive at any latitude when the meridional wavenumber is large.
In summary, we have the following findings.
(1) Waves are more likely to survive at low latitudes for strongly stratified fluid, while they can survive at any latitude for weakly stratified fluid when the meridional wavenumber is large.
(2) Waves with higher meridional wavenumber are more likely to survive.
(3) The wave frequency width is wider at lower latitudes for strongly stratified fluid. For weakly stratified fluid, the width achieves its maximum value at a certain latitude.
2.2. Phase and group velocities
Now we move back to find the solution of (2.5). Given
$\sigma$ in the frequency range
$(\sigma _{min},\sigma _{max})$, the wave solution can be obtained if (2.5) can be solved. The general solution of (2.5) can be written in a form (Gerkema & Shrira Reference Gerkema and Shrira2005) of
where
$\delta =-kB/C$ and
$\varPsi (z)=\psi (z)\exp (\textrm {i}\delta z)$. Making a substitution, we obtain
or
where
$A_{0}=\tilde {f}_{s}^2-\sigma ^2$. From this equation, we know that the wave solution depends on the structure of
$N^2(z)$.
Before the discussion of wave transmission it is necessary to consider wave directions in detail. Here we determine the wave direction by its group velocity, because it represents the direction of energy propagation. For the convenience of discussion, let us first consider a simple case with constant
$N^2$. In such case, the basis functions of wave solution can be written as
$\exp ({\textrm {i}(\pm r_{0}+\delta )z})$, where
$r=\pm r_{0}+\delta$ is the wavenumber along the vertical direction. For the convenience of discussion, we call
$\pm r_{0}$ as modified vertical wavenumbers. Substituting
$\exp ({\pm \textrm {i}r_{0}z})$ into (2.11), we obtain the dispersion relation
and its derivative
Here we define
$c_{g_{0}}=\textrm {d}\sigma /\textrm {d}r_{0}$ as the modified vertical group velocity, and
$c_{p_{0}}=\sigma /r_{0}$ as the modified vertical phase velocity.
Now we consider the vertical group velocity of the wave
$\textrm {e}^{\textrm {i}rz}$. It can be shown that
Correspondingly, we define
$c_{g}=\textrm {d}\sigma /\textrm {d}r$ as the vertical group velocity, and
$c_{p}$ as the vertical phase velocity. Note that
$c_{g}$ has the same sign as
$\textrm {d}r/\textrm {d}\sigma$. Compared with group velocity, the judgement of phase velocity direction is much easier. So here we try to build up some relations between the vertical group velocity
$c_{g}$ and the modified phase velocity
$c_{p_{0}}$. It can be shown that
Thus we conclude that
where the operator Sgn is a sign function. As a result, the sign of
$c_{p_{0}}c_{g}$ is only determined by the sign of
$C$. In other words, the group velocity and the modified phase velocity have same directions when the frequency is subinertial
$\sigma ^2 < f^2$, and opposite directions when the frequency is superinertial
$\sigma ^2>f^2$. A singularity appears when
$\sigma ^2=f^2$ and it defines the critical colatitudes
$\theta _{c}=\cos ^{-1}\pm \sigma /(2\varOmega )$ (Rieutord, Georgeot & Valdettaro Reference Rieutord, Georgeot and Valdettaro2001; Goodman & Lackner Reference Goodman and Lackner2009).
Our conclusion is general for the wave propagation on a tilted
$f$-plane. The special cases of the pure gravity wave (
$f^2=0$) and the pure inertial wave (
$N^2=0$) have been well studied in a non-tilted plane (
$\tilde {f}^2=0$ and
$c_{p_{0}}=c_{p}$) (Rieutord Reference Rieutord2015). For the pure gravity wave case (
$f^2=0$ and
$\tilde {f}^2=0$), we have
$f^2=0 < \sigma ^2 < N^2$. Thus
$c_{g}$ always points in the opposite direction to
$c_{p}$. For the pure inertial wave case (
$N^2=0$ and
$\tilde {f}^2=0$), we have
$0 < \sigma ^2 < f^2$. Thus
$c_{g}$ always propagates in the same direction as
$c_{p}$. These two results are implied in the calculations of Rieutord (Reference Rieutord2015) (see (5.48) and (8.20) in the book).
Now let us consider the gravito-inertia wave on a non-tilted f-plane (
$\tilde {f}=0$ and
$f=2\varOmega$). For such case, the lower and upper limits of wave frequency are
$\sigma ^2_{1,2}=(f^2+N^2\mp |f^2-N^2|)/2$. When the density structure is weakly stratified
$N^2/f^2 < 1$, we have
$\sigma ^2 < \sigma _{2}^2=f^2$, therefore,
$c_{g}$ has the same direction as
$c_{p}$. When the density structure is strongly stratified
$N^2/f^2>1$, we have
$f^2=\sigma _{1}^2 < \sigma ^2$, therefore,
$c_{g}$ has the opposite direction to
$c_{p}$. When
$N^2/f^2=1$, the group velocity
$c_{g}=0$ and wave energy does not propagate along the vertical direction. Clearly, the direction of
$c_{g}$ depends on the density stratification. In summary,
$c_{g}$ in weakly (strongly) stratified density structure prefers the same (opposite) direction as (to)
$c_{p}$.
Now we consider the more general case for the gravito-inertial wave on a tilted f-plane (
$\tilde {f}\neq 0$). The discussion is similar to that of the non-tilted case. If
$\tilde {f}_{s}\neq 0$, then the frequency domain
$\sigma ^2\in (\sigma _{1}^2,\sigma _{2}^2)$ can be separated into two regions by
$f^2$. In the subinertial region
$\sigma _{1}^2\in (\sigma _{1}^2,f^2)$, the group velocity
$c_{g}$ has the same direction as the modified phase velocity
$c_{p_{0}}$ (figure 1c). In contrast, in the superinertial region
$\sigma ^2\in (f^2,\sigma _{2}^2)$, the group velocity
$c_{g}$ has the opposite direction to the modified phase velocity
$c_{p_{0}}$ (figure 1d). In the strongly stratified (slowly rotating) case, the subinertial range is thinner than the superinertial range. So
$c_{g}$ and
$c_{p_{0}}$ prefer the opposite direction in a strongly stratified density structure. In a weakly stratified (rapidly rotating) case, the conclusion depends on latitudes. Both
$c_{g}$ and
$c_{p_{0}}$ prefer the same direction at high latitudes, but the opposite direction at low latitudes.
Our findings are summarised as follows.
(1) The vertical group velocity and the modified vertical phase velocity have same directions for a subinertial wave, and opposite directions for a superinertial wave.
(2) On a non-tilted plane, the vertical group and phase velocities are always in opposite directions for a pure gravity wave, and in the same directions for a pure inertial wave.
(3) On a non-tilted plane, the vertical group and phase velocities of a gravito-inertia wave are always in opposite directions in strongly stratified fluid, and the same directions in weakly stratified fluid.
2.3. Wave energy fluxes and transmission coefficients
Now we discuss the wave reflection and transmission at the interface between the convective layer and the stratified stable layer. Wei (Reference Wei2020a) has discussed four configurations on wave reflection and transmission: from convective (stable) to stable (convective) layers, and a stable layer sitting above (below) convective layer. It is easy to verify that when the stable layer is switched with the convective layer, the Boussinesq equations on the f-plane are invariant under the transformation
$z \leftrightarrow -z$ and
$w\leftrightarrow -w$, where the location of the interface is set at
$z=0$. Thus it is sufficient to discuss one of the cases. The other case can be inferred from the symmetric property. The up/down symmetry also holds under the transformation
$\theta \leftrightarrow {\rm \pi}-\theta$, so hereafter we only focus on the f-plane in the northern atmosphere. In this paper, we consider two configurations for the stable layer sitting above the convective layer: the wave propagates from the convective layer to the stable layer (figure 2a,c), and the wave propagates from the stable layer to the convective layer (figure 2b,d). In either case, we consider two different stratification structures in the stable layer: uniform stratification (figure 2a,b), and continuously varying stratification (figure 2c,d).
Sketch plots of wave propagation in different configurations. (a,c) Configuration 1: the wave propagates from the convective layer to the stable layer. (b,d) Configuration 2: the wave propagates from the stable layer to the convective layer. (a,b) The stable layer is uniform stratification. (c,d) The stable layer is continuously varying stratification.

To obtain the wave solution, we firstly discuss the general solutions in the convective layer and the stable layer, and then determine the coefficients by matching boundary conditions at the interface. We have two boundary conditions at the interface. The first is the condition for continuous vertical velocity
$w(0^{+})=w(0^{-})$. The second is the condition that Lagrangian perturbation of pressure is continuous at the interface, which requires that the first derivative of vertical velocity is continuous,
$w_{z}(0^{+})=w_{z}(0^{-})$ (see Wei (Reference Wei2020a) and Appendix B for explanation).
In the adiabatically convective layer with
$N^2=0$, it is easy to obtain the following wave solution:
where
$q>0$ and
$q^2=k^2 ({B^2-A_{0}C})/{C^2}$ is the square of wavenumber in the vertical direction. The coefficients
$a_{1}$ and
$a_{2}$ would be determined from matching boundary conditions at the interface. The wave solution in the stable layer is more complicated. It depends on the stratification structure of
$N^2(z)$. In the following, we discuss the wave solutions of two different stratification structures: uniform stratification and continuously varying stratification. Before the discussion, we introduce the vertical component of the averaged internal wave energy flux (see Appendix B for the derivation)
where the bracket
$\langle \rangle$ represents that the average is taken in a wave period, and the operator
$\textrm {Re}$ and
$\textrm {Im}$ denote the real and imaginary parts of a complex number, respectively. Given the averaged kinetic energy fluxes of the incident wave
$\left \langle F_{i}\right \rangle$, the reflective wave
$\left \langle F_{r}\right \rangle$ and the transmitted wave
$\left \langle F_{t}\right \rangle$, we define the reflection ratio
and the transmission ratio
Note that here the definitions of the reflection ratio and the transmission ratio are calculated by energy flux (Wei (Reference Wei2020a) used kinetic energy but in his corrigendum (Wei Reference Wei2020b) he used energy flux). A detailed discussion on the definitions of transmission and reflection ratios can be found in Sutherland (Reference Sutherland2010). In the following we discuss wave solutions in the two different stratifications.
3. Results
3.1. Uniform stratification in the stable layer
The case of uniform stratification has been discussed in Wei (Reference Wei2020a). Here we discuss it again in a general form for the convenience of comparison. For a uniform stratification
$N_{max}^2=\gamma _{1}$ (
$\gamma _{1}>0$), the wave solution can be written as
where
$s>0$ and
$s^2=k^2 ({B^2-A_{0}C-\gamma _{1} C})/{C^2}$ is the square of wavenumber in the vertical direction. The wave solution contains two separated branches with opposite vertical propagating directions. The selection of wave directions depends on configurations. We consider configurations 1 and 2 in the following, respectively.
3.1.1. Configuration 1
The vertical component of the group velocity (
$c_{g}$) of the transmitted wave should be in the same direction as that of the incident wave. For configuration 1,
$c_{g}$ of the transmitted wave is outward. The direction of
$c_{g}$ is determined by the sign of
$C$. Here we discuss the two cases
$C>0$ and
$C<0$, separately.
For the case
$C>0$ (or
$\sigma ^2 < f^2$), it requires
$b_{2}=0$. Waves with wavenumber
$+q$ are incident waves, the wave with wavenumber
$-q$ is the reflective wave and the wave with wavenumber
$+s$ is the transmitted wave. Matching boundary conditions at the interface gives
Note that terms containing
$\delta$ can be cancelled on both sides in the second matching boundary condition. Solving these equations, we obtain the wave amplitude ratios
With the amplitude ratios, the reflection ratio can be calculated as
and the transmission ratio can be calculated as
It can be seen that
which means that the incident wave energy is either reflected or transmitted. We are interested in whether waves could penetrate across the interface efficiently. The efficiency can be estimated by the transmission ratio (or the reflection ratio). The higher transmission ratio (or the lower reflection ratio), the more wave energy is transmitted.
Noting
$q=k\sqrt {\varDelta _{0}}/C$ and
$s=k\sqrt {\varDelta _{m}}/C$, we can rewrite the transmission ratio as
\begin{equation} \eta=1-\left(\frac{\sqrt{\varDelta_{0}}-\sqrt{\varDelta_{m}}}{\sqrt{\varDelta_{0}}+\sqrt{\varDelta_{m}}}\right)^2=1- \left(1-\frac{2}{\sqrt{\varDelta_{0}/\varDelta_{m}}+1}\right)^2,\end{equation}
where
$\varDelta _{0}=B^2-A_{0}C$ and
$\varDelta _{m}=B^2-A_{0}C-N_{max}^2C$. It can be seen that
$\eta$ decreases with
$\varDelta _{0}/\varDelta _{m}$. By taking the first derivatives, it is easy to prove that
$\varDelta _{0}/\varDelta _{m}$ increases with
$N_{max}^2/(2\varOmega )^2$ and decreases with
$\sigma ^2$. As a result, we find that
$\eta$ decreases with
$N_{max}^2/(2\varOmega )^2$ and increases with
$\sigma ^2$. Specifically, we have
$\eta \rightarrow 1$ when
$N_{max}^2/(2\varOmega )^2\rightarrow 0$ or
$\sigma ^2\rightarrow f^2$. Therefore, the transmission is efficient when the rotation is fast, or when the absolute value of the frequency is close to the vertical Coriolis parameter.
For the case
$C<0$ (or
$\sigma ^2>f^2$), it requires
$b_{1}=0$. The wave with wavenumber
$-q$ is the incident wave, the wave with wavenumber
$+q$ is the reflective wave and the wave with wavenumber
$-s$ is the transmitted wave. Following the same procedure, we find that the transmission ratio is the same as (3.9). Noting
$\varDelta _{0}/\varDelta _{m}<1$, we see that
$\eta$ increases with
$\varDelta _{0}/\varDelta _{m}$. By taking derivatives, we find that
$\varDelta _{0}/\varDelta _{m}$ decreases with
$N_{max}^2$ and
$\sigma ^2$. As a result, we find that
$\eta$ decreases with
$N_{max}^2/(2\varOmega )^2$ and
$\sigma ^2$. Again, we conclude that the transmission ratio is efficient when the stable layer is weakly stratified, or when the wave is near the critical colatitude.
For the configuration 1, we have plotted the transmission ratios for different
$N_{max}^2/(2\varOmega )^2$ and
$\sin ^2\alpha$ in figure 3. The effect of
$N_{max}^2/(2\varOmega )^2$ is shown from the top to the bottom (
$N_{max}^2/(2\varOmega )^2=10,1,0.1$), and the effect of
$\sin ^2\alpha$ is shown from the left to the right (
$\sin ^2\alpha =0.01,0.5,1.0$). First, we observe that the degree of stratification (
$N_{max}^2/(2\varOmega )^2$) has a great impact on the frequency domain. The frequency domain is much wider at low (high) latitudes when the stable layer is strongly (weakly) stratified. Second, we notice that the wavenumber also has an important effect on the width of the frequency domain. The frequency domain is much wider when the meridional wavenumber dominates the zonal wavenumber. These two points have already been confirmed in the frequency domain analysis. Third, we notice that the transmission ratio largely depends on the degree of stratification and wave frequency. High transmission ratios are achieved when
$N_{max}^2/(2\varOmega ^2)$ is small (the stable layer is weakly stratified), or when
$\sigma ^2-f^2\simeq 0$ (the wave is near the critical colatitude). This result is consistent with our theoretical analysis. In the study of wave transmission in a multilayered structure, André, Barker & Mathis (Reference André, Barker and Mathis2017) also found that low-frequency waves of all wavelengths are able to perfectly transmit near the critical colatitude. The reason is that only one non-zero solution of
$c_{g}$ exists at the critical colatitude, and thus the amplitude of the incident wave must be equal to that of the transmitted wave at the boundary (André et al. Reference André, Barker and Mathis2017).
Transmission ratios for different
$N_{max}^2/(2\varOmega )^2$ and
$\sin ^2\alpha$. The horizontal axis is
$\cos ^2\theta$ and the vertical axis is
$\sigma ^2/(2\varOmega )^2$. (a–c) The slowly rotating group (
$N_{max}^2/(2\varOmega )^2=10$) with
$\sin ^2\alpha \in \{0.01,0.5,1.0\}$ from the left to right. (d–f) The moderately rotating group (
$N_{max}^2/(2\varOmega )^2=1$) with
$\sin ^2\alpha \in \{0.01,0.5,1.0\}$ from the left to right. (g–i) The rapidly rotating group (
$N_{max}^2/(2\varOmega )^2=0.1$) with
$\sin ^2\alpha \in \{0.01,0.5,1.0\}$ from the left to right. Waves can only survive in the coloured regions. The regions are left white in the contour plots if waves cannot survive.

3.1.2. Configuration 2
For configuration 2, the wave propagates from the stable layer to the convective layer. We discuss it in two cases,
$C>0$ and
$C<0$, respectively. For the first case
$C>0$, it requires
$a_{1}=0$. Waves with wavenumber
$-s$,
$+s$ and
$-q$ are the incident wave, reflective wave and transmitted wave, respectively. From boundary conditions, we obtain
from which we obtain the transmission ratio
\begin{equation} \eta=\frac{q}{s}\left(\frac{a_{2}}{b_{2}}\right)^2=\frac{4sq}{(s+q)^2}=1- \left(\frac{\sqrt{\varDelta_{m}}-\sqrt{\varDelta_{0}}}{\sqrt{\varDelta_{m}}+\sqrt{\varDelta_{0}}}\right)^2. \end{equation}The analysis is similar to configuration 1, and the conclusion is the same.
For the case
$C<0$, we can show that the conclusion remains unchanged, and here we will not repeat it.
3.2. Continuously varying stratification in the stable layer
3.2.1. Transmission at the interface
For a continuously varying stratification in the stable layer, we assume that
$N^2(z)$ is a function of
$z$ in the region
$z\in (0,z_{m})$. For the existence of a wave solution, it requires
$B^2-A_{0}C-N^2(z)C>0$ for
$z\in (0,z_{m})$. It also requires
$N^2(0)=0$ because
$N^2(z)$ is assumed to be continuous at the interface. In the stable layer, we write (2.12) as
where
$s^2=k^2[(B^2-A_{0}C-N^2C)/{C^2}]$ and
$s$ is a function of
$z$. We perform Wentzel–Kramers–Brillouin (WKB) analysis to solve this equation. Assuming
$\psi \propto e^{iX(z)}$ and substituting it into the above equation, we obtain
It is a nonlinear equation which is generally difficult to solve. However, an approximate solution can be obtained if
$X(z)$ is assumed to be a slowly varying function. If
$X(z)$ is a slowly varying function, then it is reasonable to assume that
$X_{zz}$ is small compared with
$X_{z}^2$. Neglecting
$X_{zz}$ in (3.14) gives the first approximation
which has the solution
From a comparison of
$X_{zz}$ and
$X_{z}^2$, we find that the assumption on a slowly varying function is valid as long as
$|s_{z}/s^2| \ll 1$. In other words, it requires that the wavelength (
$1/s$) is much smaller than the variation scale length of the wavenumber (
$\partial z/\partial {\log s}$). Substitution of (3.16) into (3.14) yields the following equation:
from which we obtain the second approximation
Integration of the above expression gives
Substituting
$X$ back into
$\psi$, we obtain an approximate solution of (3.13),
Thus the general solution of
$\psi$ can be written as
and its derivative is
\begin{align} \psi_{z}&={-}2^{{-}1} s_{z}(z) s(z)^{{-}1} \psi\nonumber\\ &\quad + \textrm{i}[s(z)s(0)]^{1/2}\left[ b_{1} \exp\left({\textrm{i}\int_{0}^{z} s(z') \,\textrm{d}z'}\right) - b_{2} \exp\left({-\textrm{i}\int_{0}^{z} s(z') \,\textrm{d}z'}\right)\right]. \end{align}
At
$z=0$, we have
where
$\beta =(2s(0))^{-2}s_{z}(0)$ and
$s(0)=q$ have been used. Here the explicit form of
$\beta$ is
which depends on the slope of
$N^2$. As
$\beta$ can be expressed as
$(1/s)/(\partial {z}/\partial \log s)|_{z=0}$, it measures the importance of the wavelength to the variation scale length of the wavenumber at the interface. Here
$\beta$ is small when the wavenumber varies slowly near the interface. Later, we will show that
$\beta$ plays a critical role in wave transmissions at the interface. Now we deduce wave transmission ratios at the interface for the case
$C>0$ in configuration 1. For
$C>0$, the outward transmitted wave requires
$b_{2}=0$, and thus, the following equations can be deduced from the boundary conditions:
The solutions of the above equations are
from which we can deduce the transmission ratio
Obviously, the efficiency of transmission mainly depends on the value of
$|\beta |$. If
$|\beta | \rightarrow \infty$, then
$\eta \rightarrow 0$ and the incident wave is totally reflected. If
$|\beta |\rightarrow 0$, then
$\eta \rightarrow 1$ and the incident wave is totally transmitted. If
$\infty >|\beta |>0$, then
$1>\eta >0$ and the incident wave is partially reflected.
Let us further assume that
$N^2$ varies as an algebraic function
$N^2(z) =\gamma _{2} z^{\nu }$, where
$\gamma _{2}>0$ and
$\nu \geqslant 1$ are constants. Making a substitution, we obtain
From the above equation, we have the following conclusions.
(1) When
$\nu >1$, we have
$|\beta | =0$, thus the incident wave is totally transmitted at the interface.(2) When
$\nu =1$, we have
$\infty >|\beta |>0$, thus the incident wave is partially transmitted at the interface.
It has to be mentioned that above conclusions are drawn based on the WKB analysis, which generally requires
$|\beta |=|(2s(0))^{-2}s_{z}(0)|\ll 1$. Under this condition, wave transmission is expected to be efficient. Although (3.30) is obtained under the WKB approximations, in the Appendix C we have shown in a non-WKB analysis that it probably still holds for large
$|\beta |$.
Let us now consider the special case
$\nu =1$. For this case, we have
where the relation
$N_{max}^2=\gamma _{2}z_{m}$ has been applied. It can be seen that
$|\beta |$ decreases with
$z_{m}k$ and increases with
$N_{max}^2/(2\varOmega )^2$. Also, it can be proved that
$|\beta |$ decreases with
$\sigma ^2$ by taking derivatives. As a result,
$\eta$ increases with
$z_{m}k$ and
$\sigma ^2$, and decreases with
$N_{max}^2/(2\varOmega )^2$. Specifically, we have
$\eta \rightarrow 1$ when
$N_{max}^2/(2\varOmega )^2 \rightarrow 0$, or
$\sigma ^2 \rightarrow f^2$, or
$z_{m}\gg k^{-1}$. Therefore, the transmission is efficient when the stable layer is weakly stratified, or when the wave is near the critical latitude, or when the thickness of the linearly varying stratification layer is far greater than the horizontal wavelength. Figure 4 illustrates the transmission contours at different
$z_{m}k$ and
$N_{max}^2/(2\varOmega )^2$. It clearly shows that the transmission ratio increases with
$z_{m}k$ even when the stable layer is strongly stratified. A large value of
$z_{m}k$ can be easily realised for short waves. It indicates that short waves are more likely to transmit in a strongly stratified fluid when the stratification in the stable layer is linearly varying.
Transmission ratios at different
$z_{m}k$ (
$=0.1,1,10$ from the left to right) and
$N_{max}^2/(2\varOmega )^2$ (
$=10,1,0.1$ from the top to the bottom) when the stratification is linearly varying
$N^2=\gamma _{2}z$. The parameter
$\sin ^2\alpha =1.0$ for all the shown cases. Waves can only survive in the coloured regions. The regions are left white in the contour plots if waves cannot survive.

Following similar procedures, we find that the transmission ratios remain unchanged for
$C<0$, and for configuration 2. Here we will not repeat the discussion for these cases.
3.2.2. Far-field transmission away from the interface
In previous sections, we mainly consider wave transmissions at the interface. By the WKB analysis, we have found that the incident wave can be totally transmitted at the interface for a convexly varying stratification. However, wave reflection still can occur in the far-field away from the interface. In this section, we will discuss the far-field wave transmission for linearly and convexly varying buoyancy frequency profiles (
$\nu \geqslant 1$) through WKB analysis. We consider a three-layer structure, with a bottom convective layer, a middle transition stable layer with a algebraic function of
$N^2=\gamma _{2} z^{\nu }$ and a top stable layer with constant
$N^2=N_{max}^2$. The bottom, middle and top layers are separated at
$z=0$ and
$z=z_{m}$, respectively. Here we only consider the case with
$C>0$. The discussion on the case with
$C<0$ is similar, and the conclusion remains unchanged. Let
$a_{1}$ and
$a_{2}$ be the amplitudes of incident and reflective waves at the bottom layer,
$b_{1}$ and
$b_{2}$ be the amplitudes of upward and downward propagating waves in the middle layer and
$b_{3}$ be the transmitted wave at the top layer. Matching the boundary conditions at the two interfaces, we can obtain the relations among these amplitudes. After some trivial calculations, we find that the far-field transmission can be written as
where
$\beta _{m}=(2s(z_{m}))^{-2}s_{z}(z_{m})$, and
$\chi$ is the argument of
$\exp (-\textrm {i}\int _{0}^{z_{m}}2s(z')\,\textrm {d}z')$. When
$\nu >1$, it is easy to obtain
with
Interestingly, it can be shown that the far-field transmission ratio for the cases
$\nu >1$ also increases with
$z_{m}k$ and decreases with
$N_{max}^2/(2\varOmega )^2$, which is quite similar to the result obtained at the interface for the case
$\nu =1$.
The WKB analysis is not valid for the cases
$0<\nu <1$. We will discuss these cases in the following section by numerical calculations. In summary, for far-field transmission, we mainly have the following conclusion.
(1) For
$\nu >1$, although the incident wave is totally transmitted at the interface, it is partially transmitted at the far-field away the interface.(2) For
$\nu >1$, the far-field transmission can be efficient when the thickness of the stratification layer is far greater than the horizontal wavelength, but less efficient the other way around.
3.3. Numerical solutions of transmission ratios for different
$\nu$
In the above WKB and non-WKB analysis, we have considered the transmission ratio for
$N^2=\gamma _{2}z^{\nu }$ with
$\nu \geqslant 1$. The case
$0<\nu <1$ has not been discussed yet. When
$0<\nu <1$, the WKB approximation is not valid. For this case, we try to solve (2.12) numerically. However, solving (2.12) directly is not easy for the current configurations, because the wavenumber of the outgoing wave is not well defined. As a result, the wavenumber of the transmitted wave has to be estimated. This estimation may affect the transmission ratio at the radiative–convective boundary. To gain some insights, we estimate the wavenumber of the transmitted wave at an arbitrary points
$0 < z_{b} < z_{m}$, and then investigate the effects on the transmission ratio. At the location
$z=z_{b}$, we estimate the wavenumber
$s_{z_{b}}$ of the transmitted wave at
$z=z_{b}$ as
where the coefficients
$c_{1}$ and
$c_{2}$ are
As
$s_{z_{b}}^2$ is a constant, wave directions can be easily separated, and the proper transmitted wave can be selected. Given boundary conditions at
$z=z_{b}$, (2.12) can be integrated backward to the interface. Once
$\psi (0)$ and
$\psi _{z}(0)$ are found numerically, we can calculate the amplitude ratio of the reflected wave to the incident wave, from which the reflection and transmission ratios can be deduced as well.
Here we show transmission ratios for subinertial (
$\sigma ^2 < f^2$) and superinertial waves (
$\sigma ^2>f^2$), respectively. For subinertial waves, both
$c_{1}$ and
$c_{2}$ are positive values; for superinertial waves, however, both
$c_{1}$ and
$c_{2}$ are negative values (see the § 3.4 for further discussions). Figure 5 shows transmission ratios at different values of
$\nu$,
$c_{1}$ and
$c_{2}$ for subinertial waves. The
$x$-axis is the relative position of the boundary. As the wavenumber of the transmitted wave is approximated, we note that numerical results do not have to be completely consistent with our previous WKB and non-WKB analysis. Our intention here is to gain some insights by comparative studies. From the figure, we see the following trends. First, transmissions are always efficient when
$c_{2}$ is large. A large value of
$c_{2}$ can be achieved when
$N_{max}^2/(2\varOmega )^2$ or
$f^2-\sigma ^2$ is small, which corresponds to situations at fast rotation or at critical latitudes. Second, the far-field transmission ratio (the transmission ratio at the location
$z_{b}/z_{m}=1$) increases with
$z_{m}k$ for any
$\nu$. The effect of
$z_{m}k$ on far-field transmission ratio can be investigated by keeping
$c_{2}$ a constant and varying
$c_{1}$ (that is, keeping
$z_{m}$ a constant and varying
$k$). From figure 5, we see that the far-field transmission ratio has an increasing trend with increasing
$z_{m}k$ for any
$\nu$. The results for superinertial waves are quite similar to those obtained in subinertial waves (see figure 6), and we shall not repeat the discussion here. Therefore, for a given wave (fixed
$k$ and
$\sigma$), the variation rate of stratification (
$N_{max}^2/z_{m}$ for a fixed
$N_{max}^2$) has an important effect on the efficiency of transmission. The far-field transmission is more efficient when the variation of stratification is slower.
Transmission ratio as a function of the relative position of the chosen boundary for subinertial waves. Different combinations of
$c_{1}=0.1,1,10$ and
$c_{2}=1.1,2,10$ are shown in the panels. In each panel, transmission ratios at different
$\nu =0.1,0.5,1,2,5$ are shown with coloured lines.

Transmission ratio as a function of the relative position of the chosen boundary for superinertial waves. Different combinations of
$c_{1}=-0.1,-1,-10$ and
$c_{2}=-0.1,-1,-10$ are shown in the panels. In each panel, transmission ratios at different
$\nu =0.1,0.5,1,2,5$ are shown with coloured lines.

3.4. Non-traditional effects
Traditional approximation, in which the horizontal Coriolis parameter
$\tilde {f}=0$, is often used in geophysical fluids. Under the traditional approximation, the frequency range for wave propagation is
The minimum value of
$N_{min}=0$ is achieved in the convection zone, thus the wave frequency must be subinertial
$N_{max}^2 < \sigma ^2 < f^2$. Obviously, no inertial or gravito-inertia wave can survive in both convective and stable layers if the stable layer is strongly stratified with
$N_{max}^2>f^2$. However, the case is different when non-traditional effects are present. With non-traditional effects, figure 3(a–c) clearly show that waves can propagate in both convective and stable layers, although the stable layer is strongly stratified.
In the traditional approximation, when buoyancy frequency profile
$N^2(z)=\gamma _{2}z^{\nu }$ is assumed, superinertial waves in the stable layer are always kept away at distances
$z>(\sigma ^2/\gamma )^{1/\nu }$ from the interface, so as to ensure
$N^2>\sigma ^2 >f^2$. Wave rays propagating toward the interface are reflected back to the stable layer, without ever reaching the interface. This situation is similar to the antiwaveguide sound propagation adjacent to water surface (Brekhovskikh, Lysanov & Lysanov Reference Brekhovskikh, Lysanov and Lysanov2003). On the other hand, subinertial waves are confined next to the interface at distances
$z<(\sigma ^2/\gamma )^{1/\nu }$, so as to ensure
$N^2 < \sigma ^2 < f^2$.
When non-traditional effects present, the condition for wave propagation becomes (2.6), which can be reorganised as
for subinertial waves
$\sigma ^2 < f^2$, and
for superinertial waves
$\sigma ^2>f^2$. Subinertial waves will be confined next to the interface at distances
\begin{equation} z<\left(\frac{\sigma^2(f^2+\tilde{f}_{s}^2-\sigma^2)}{\gamma_{2}(f^2-\sigma^2)}\right)^{1/\nu}, \end{equation}
which is similar to that obtained in traditional approximation. However, the situation for superinertial waves is different. For superinertial waves with
$\sigma ^2>f^2+\tilde {f}_{s}^2$, waves are always kept away from the interface at distances
\begin{equation} z>\left(\frac{\sigma^2(\sigma^2-f^2-\tilde{f}_{s}^2)}{\gamma_{2}(\sigma^2-f^2)}\right)^{1/\nu}, \end{equation}
while for superinertial waves with
$f^2 < \sigma ^2 < f^2+\tilde {f}_{s}^2$, no similar restriction is obtained and waves can freely propagate in the stable layer. Thus, subinertial waves can only propagate a limited distance, but superinertial waves are able to propagate far away from the interface.
To further illustrate the differences, we show wave transmission ratios at the interface in the traditional approximation for
$N^2=\gamma _{2}z$ in figure 7. In the traditional approximation, only frequencies in
$(2\varOmega )^2\geqslant f^2>\sigma ^2>N^2$ can survive in both convective and stable layers. Therefore, on consideration of the transmission problem, wave frequencies must be subrotational with
$\sigma ^2 < (2\varOmega )^2$ (see also figure 11 and their explanations in Gerkema et al. (Reference Gerkema, Zimmerman, Maas and Van Haren2008)). As
$N^2$ must be smaller than
$f^2$, here we only show the weakly stratified cases with
$N_{max}^2/(2\varOmega )^2=0.1$. It is obvious that frequencies are confined in the region
$f^2>\sigma ^2>N^2$. One consequence is that waves could hardly propagate at low latitudes. However, when non-traditional effects are included, from figure 4, we see that wave propagation frequency ranges are much wider. This is especially true for waves propagating at low latitudes.
Wave transmissions under the traditional approximation. Transmission ratios computed at different
$z_{m}k$ (
$=0.1,1,10$ from the left to right) when the stratification is linearly varying
$N^2=\gamma _{2}z$. In all the three cases, the parameter
$N_{max}^2/(2\varOmega )^2=0.1$. Waves can only survive in the coloured regions. The regions are left white in the contour plots if waves cannot survive.

In the study of the tidal response of a massive rotating star, Savonije, Papaloizou & Alberts (Reference Savonije, Papaloizou and Alberts1995) found that subrotational responses with traditional approximation are confined in a poleward region, while without traditional approximation they can be found from equator to pole. The result is consistent with our analysis. This phenomenon is also reviewed in Gerkema et al. (Reference Gerkema, Zimmerman, Maas and Van Haren2008). In the study of tidal oscillations of a rotating star with 1.5 solar masses, Dintrans & Rieutord (Reference Dintrans and Rieutord2000) used a stratification profile which is assumed to be proportional to radius. This setting is similar to the linearly varying stratification discussed in our paper. In their cases, the star has a strongly stratified middle stable layer surrounded by inner and outer convective layers. In a subrotational case (
$\sigma /(2\varOmega )=2/3$), they found that characteristic trajectories associated with
$H_{2}$-mode (
$f>\sigma >0$) propagate both in the convective and radiative layers, while those associated with
$E_{1}$-mode (
$N_{max}>\sigma >f$) remain trapped in the convective layer (see figure 6a in Dintrans & Rieutord (Reference Dintrans and Rieutord2000)). Their figure also shows that
$H_{2}$-mode waves are mainly trapped at low latitudes with the colatitude
$\theta \gtrsim \arccos (2/3)\approx 48^{\circ }$. Since
$H_{2}$-mode can survive both in the convective and radiative layers, their result can be readily compared with our calculation in figure 4(a–c). As shown in the figure, when
$z_{m}k$ is large (see figure 4c), wave transmissions can be efficiently transmitted at low latitudes
$\theta <\theta _{c}$ (the diagonal line from the lower left to the upper right in the figure is
$\theta =\theta _{c}$). Our calculation shows good agreement with their result. If a traditional approximation is made, however,
$H_{2}$-mode can never survive both in convective and radiative layers when
$N_{max}^2>f^2$.
4. Conclusions and discussion
We have discussed the wave propagation in a partially stratified rotating fluid with two configurations. In configuration 1, the wave propagates from the convective layer to the stable layer. In configuration 2, the wave propagates from the stable layer to the convective layer. First, we have discussed the width of the frequency range on the existence of a wave solution. We have the following major findings.
(1) When the stable layer is strongly stratified (slow rotation), there is a preference for waves to survive at low latitudes rather than high latitudes.
(2) When the stable layer is weakly stratified (rapid rotation), waves can survive at any latitude if the meridional wavenumber is large.
Second, we have discussed the relationship between the vertical phase and group velocities of a wave in the rotating fluid. We have the following major findings.
(1) On a general f-plane, the vertical group velocity and the modified vertical phase velocity have same directions for a subinertial wave, and opposite directions for a superinertial wave.
(2) On a non-tilted plane, the vertical group and phase velocities of a gravito-inertia wave are always in opposite directions in strongly stratified fluid, and same directions in weakly stratified fluid.
Third, we have investigated the efficiency of wave transmission in these configurations. We have not observed significant differences on the efficiency of wave transmission between these two configurations. We have considered different stratification structures in the stable layer: the uniform stratification, and the continuously varying stratification. For the continuously varying stratification when the square of buoyancy frequency
$N^2$ varies as an algebraic function
$N^2 \propto z^{\nu }$ (
$\nu \geqslant 1$), we have discussed the transmission ratio both analytically (for
$\nu \geqslant 1$) and numerically (for any
$\nu >0$). We have the following major findings.
(1) For the uniform stratification, the transmission at the interface is efficient when the stable layer is weakly stratified (rapid rotation), or when the wave is at the critical latitude.
(2) For the continuously varying stratification (
$N^2 \propto z^{\nu }$ with
$\nu >0$), the far-field transmission is efficient when the stable layer is weakly stratified (rapid rotation), or when the wave is at the critical latitude, or when the thickness of the stratification layer is far greater than the horizontal wavelength (it means that
$N^2$ varies slowly from 0 to
$N_{max}^2$ at a distance much greater than the horizontal wavelength).
For each stratification structure, it is also interesting to note that the wave transmission ratio at the interface is identical for both configurations, no matter what direction the wave propagates. It only depends on the characteristics of the wave (wave frequency and wavenumber) and the fluid (degree of stratification). Our findings have useful applications in the prediction of wave transmission in rotating stars or planets. Our calculation shows that wave transmission largely depends on the stratification structure. Wave transmissions are more efficient when the transition layer of radiative–convective varies slower. Figure 8 shows typical structures of buoyancy frequency squared in two main sequence stars. The blue curve denotes the Sun, and the orange curve is a massive star with three times solar masses. For this massive star, it can be seen that the transition of buoyancy frequency squared is almost linear; for the Sun, the transition is more abrupt. Thus we expect that the wave transmission is more efficient in this massive star.
Brunt–Väisälä frequency as a function of radius. The blue curve shows the solar case, and the orange curve shows the case of a main-sequence massive star with three times solar masses. The structures are computed with the MESA code (Paxton et al. Reference Paxton, Bildsten, Dotter, Herwig, Lesaffre and Timmes2010).

Linear instability analysis of rotating viscous flow has shown that teleconvection (fluid motion is vigorous in the stable layer) can be driven when the stable layer is weakly stratified in spherical geometry (Zhang & Schubert Reference Zhang and Schubert2002). This phenomenon has also been verified in the analysis of rapidly rotating viscous flow on f-planes (Cai Reference Cai2020). These results have borne some similarities with our conclusions. So, probably, our arguments still hold in other geometries; it would be interesting to verify them in different geometries. It has to be mentioned that our analysis is based on local f-planes in Cartesian coordinates, in which the curvature effect has not been included. New types of waves, such as Rossby and Poincaré waves, can propagate if the variation of the Coriolis parameter on latitude is taken into consideration. Also, f-plane only focuses on local waves with short wavelengths. In a global sphere, transmission of global waves may differ significantly from that of local waves (Pontin et al. Reference Pontin, Barker, Hollerbach, André and Mathis2020). Future extensions may include considering wave transmissions in beta planes or a global sphere. Many stars and planets have multiple layer structures. Sutherland (Reference Sutherland2010) discussed transmission of gravity waves into neighbouring neutrally stable regions. André et al. (Reference André, Barker and Mathis2017) discussed transmission of internal waves in layered structures. Gerkema & Shrira (Reference Gerkema and Shrira2005) considered multiple waveguides for waves in the ocean. Wave transmissions with varying stratification in a layered structure would be an interesting question for future research.
Acknowledgements
We thank three anonymous reviewers for their insightful comments and suggestions on this manuscript.
Funding
T.C. has been supported by NSFC (no. 11503097), the Guangdong Basic and Applied Basic Research Foundation (no. 2019A1515011625), the Science and Technology Program of Guangzhou (no. 201707010006), the Science and Technology Development Fund, Macau SAR (nos. 0045/2018/AFJ, 0156/2019/A3) and the China Space Agency Project (no. D020303). C.Y. has been supported by the National Natural Science Foundation of China (grants 11373064, 11521303, 11733010, 11873103), Yun-nan National Science Foundation (grant 2014HB048) and Yunnan Province (2017HC018). X.W. has been supported by National Natural Science Foundation of China (grant no. 11872246) and Beijing Natural Science Foundation (grant no. 1202015). This work is partially supported by Open Projects Funding of the State Key Laboratory of Lunar and Planetary Sciences.
Declaration of interests
The authors report no conflict of interest.
Appendix A
In this appendix, we analyse the dependence of the width of frequency range
$\sigma _{max}^{2}-\sigma _{min}^2$ on the colatitude
$\theta$, the plane wave propagating direction
$\alpha$ and the degree of stratification
$N_{max}^2/(2\varOmega )^2$. The width of the frequency range (2.9) can be simplified as
where
$G(\theta ,\mu _{1},\mu _{2})=\sigma _{max}^2-\sigma _{min}^2$,
$G_{1}(\theta ,\mu _{1},\mu _{2})=\cos ^2\theta +\mu _{1}\sin ^2\theta -\mu _{2}$,
$G_{2}(\theta ,\mu _{1}, \mu _{2})=\mu _{1}\mu _{2}\sin ^2\theta$,
$\mu _{1}=\sin ^2 \alpha$ and
$\mu _{2}=N^2_{max}/(2\varOmega )^2$.
We first investigate the monotonicity of
$\sigma _{max}^2-\sigma _{min}^2$ on
$\theta$. Taking the first derivative with respect to
$\theta$ and setting
$G_{\theta }(\theta )=0$, we obtain
From the above equation, we obtain that the critical point should satisfy
and
$G_{\theta }$ has the same sign as
$\mu _{2}-(1-\mu _{1})$. If
$\mu _{2}>1$, then the frequency width increases with
$\theta$. If
$\mu _{2}<1$, there exists a critical value
$\mu _{1c}=1-\mu _{2}$, and the frequency width increases (decreases) with
$\theta$ when
$\mu _{1}>\mu _{1c}$ (
$\mu _{1}<\mu _{1c}$). Figures 9(a) and 9(b) present two cases with
$\mu _{2}<1$ and
$\mu _{2}>1$, respectively. Figure 9(b) clearly shows that
$G(\theta )$ increases with
$\theta$ when
$\mu _{2}>1$. Figure 9(a) shows that the trend of
$G(\theta )$ depends on a critical value
$\mu _{1c}$ when
$\mu _{2}<1$.
(a) Here
$G(\theta )$ as a function of
$\theta$ when
$\mu _{2}=2/3$; in this case,
$\mu _{2}<1$ and
$\mu _{1c}=1/3$. (b) Here
$G(\theta )$ as a function of
$\theta$ when
$\mu _{2}=3/2$; in this case,
$\mu _{2}>1$.

Then we consider the monotonicity of
$\sigma _{max}^2-\sigma _{min}^2$ on
$\alpha$. Taking the first derivative of
$G$ on
$\mu _{1}$, we have
\begin{equation} G_{\mu_{1}}=\left(1+\frac{\cos^2\theta+\mu_{1}\sin^2\theta+\mu_{2}}{\sqrt{G_{1}^2 +4G_{2}}}\right)\sin^2\theta\geqslant 0. \end{equation}
Therefore, the frequency width increases with
$\sin ^2\alpha$.
Last, we consider the monotonicity of
$\sigma _{max}^2-\sigma _{min}^2$ on the degree of stratification. The first derivative of G on
$\mu _{2}$ gives
\begin{equation} G_{\mu_{2}}=\frac{-4\mu_{1}\sin^2\theta \cos^2\theta}{(\sqrt{G_{1}^2 +4G_{2}})(\mu_{1}\sin^2\theta+\mu_{2}-\cos^2\theta+\sqrt{G_{1}^2 +4G_{2}})}\leqslant 0. \end{equation}
Therefore, the frequency width decreases with
$N_{max}^2/(2\varOmega )^2$.
Appendix B
In this appendix, we deduce the averaged internal wave energy flux, which is needed in the discussion of wave transmission and reflection. Applying the operation
$\boldsymbol {\nabla } \times$ to (2.2), we obtain the vorticity equation
From this equation, we can obtain the vertical component of vorticity
Solving (2.3) and (B2), we get the horizontal velocities
from which, we can also obtain the modified pressure
$p$ as follows:
\begin{align} p&=\frac{1}{\textrm{i}k\sin\alpha}(\textrm{i}\sigma v-fu)\nonumber\\ &=\frac{[(f^2-\sigma^2) \,\textrm{Im}(w_{z}/w)+k\widetilde{f_{s}}f]+\textrm{i} [k\tilde{f}\sigma\cos\alpha-(f^2-\sigma^2)\,\textrm{Re}(w_{z}/w)]}{k^2\sigma}w. \end{align}
Equation (B5) indicates that continuous pressure at the interface requires that the first derivative of vertical velocity to be continuous. Now we consider the energy equation. Computing
$\boldsymbol {u}\, \cdot$ (2.2)
$+\, b/N^2\, \times$ (2.3), we can obtain the energy equation
where
$E_{kin}=|\boldsymbol {u}|^2/2$ is the kinetic energy,
$E_{pe}=b^2/(2N^2)$ is the potential energy and
$E_{wave}=E_{kin}+E_{pe}$ is the wave energy. From (B3) and (B4) and
$b=-N^2 w/(\textrm {i}\sigma )$, we can calculate the averaged wave energy
\begin{align} \left\langle E_{wave}\right\rangle&=\left\langle E_{kin}\right\rangle+\left\langle E_{pe}\right\rangle\nonumber\\ &=\frac{1}{4\sigma^2 k^2}[|f\varPsi_{z}/\varPsi+\textrm{i}k\tilde{f}_{s}|^2+\sigma^2(|\varPsi_{z}/\varPsi|^2+ k^2)+N^2 k^2]|\psi|^2 \end{align}and the vertical component of the averaged internal wave energy flux
\begin{align} \left\langle F\right\rangle&=\left\langle\textrm{Re}(p)\,\textrm{Re}(w)\right\rangle\nonumber\\ &=\left\langle\frac{[(f^2-\sigma^2) \,\textrm{Im}(w_{z}/w)+k\widetilde{f_{s}}f] \,\textrm{Re}(w)\,\textrm{Re}(w)}{k^2\sigma}\right\rangle\nonumber\\ &=\frac{(f^2-\sigma^2) \,\textrm{Im}(\varPsi_{z}/\varPsi)+k\widetilde{f_{s}}f}{2k^2\sigma}|\varPsi|^2\nonumber\\ &=\frac{C \delta + C\,\textrm{Im}(\psi_{z}/\psi)+kB}{2k^2\sigma}|\psi|^2\nonumber\\ &=\frac{C\,\textrm{Im}(\psi_{z}/\psi)}{2k^2\sigma}|\psi|^2. \end{align}Appendix C
We have considered the continuously varying stratification in the stable layer under the WKB approximations. For two special cases with
$\nu =1$ and
$\nu =2$, exact solutions can be found without additional approximations. For these two special cases, we attempt to solve (2.12) directly to verify the validity of the WKB results.
C.1. Solution for linearly varying
$N^2$
With a linear relation
$N^2=\gamma _{2} z$, (2.12) can be written as
Making a substitution
$z'=\textrm {Sgn}(C)\ell ^{2/3}(z-z_{t})$, we can transform it into a standard form,
where
$\ell ^2=k^2\gamma _{2}/|C|$ and
$z_{t}=(B^2-A_{0}C)/(\gamma _{2}C)$. This equation is a Sturm–Liouville problem and its solution can be found by using Airy functions,
where Ai and Bi are the first and second kinds of Airy functions. It has been found that the Airy functions have the following asymptotic expansions for
$z'\ll -1$ (Zhang & Lai Reference Zhang and Lai2006; Olver et al. Reference Olver, Lozier, Boisvert and Clark2010):
where
$X(z')=2/3(-z')^{3/2}+{\rm \pi} /4$. At the interface, we have
where
$z_{0}=\textrm {Sgn}(C)z_{t}>0$,
${\rm \pi} _{0}={\rm \pi} ^{-1/2}\ell ^{-1/6}z_{0}^{-1/4}$,
$X_{0}=X(-\ell ^{2/3}z_{0})$, and the relation
$q=\ell z_{0}^{1/2}$ is used. After some trivial calculations, we find that the transmission ratio can be evaluated as
where
Apparently,
$\eta$ increases with
$z_{0}q$. Comparing (C8) with (3.30), we can see that the following relation holds:
C.2. Solution for quadratically varying
$N^2$
For the quadratic stratification in the stable layer, we assume the profile is
$N^2=\gamma _{2}z^2$. Equation (2.12) is
Substituting
$z'=(4\gamma _{2}k^2/|C|)^{1/4}z$, we can transform the equation to a standard form of Sturm–Liouville problem,
where
$\lambda =k(B^2-A_{0}C)/(2\gamma _{2}^{1/2}|C|^{3/2})$. The solution of the above equation can be found by using the parabolic cylinder functions (Olver et al. Reference Olver, Lozier, Boisvert and Clark2010).
For
$C>0$, the problem is
and the solution is
where
$U$ and
$V$ are Weber parabolic cylinder functions. Asymptotic expansions of
$U(-\lambda ,z')$ and
$V(-\lambda ,z')$ can be found when
$\lambda$ is large and
$0\leqslant z'<2\sqrt {\lambda }$ (Olver et al. Reference Olver, Lozier, Boisvert and Clark2010),
where
$\mu =\sqrt {2}\lambda ^{1/2}$,
$\tilde {z}=z'/(\sqrt {2}\mu )$,
$\epsilon =1/\varGamma (\lambda +1/2)$,
$\varGamma (\xi )$ is the Gamma function of
$\xi$ and
$X(\tilde {z})=\lambda (\arccos \tilde {z}-\tilde {z}\sqrt {1-\tilde {z}^2})+\frac {1}{4}{\rm \pi}$. At the interface, we have
where
$X_{0}=(2\lambda +1){\rm \pi} /4$.
For
$C<0$, the problem is
and its solution is
where
$W$ is another type of Weber parabolic cylinder function (Olver et al. Reference Olver, Lozier, Boisvert and Clark2010). The asymptotic solution of
$W$ in
$0\leqslant z'<2\sqrt {\lambda }$ can be found as
where
$Y(\tilde {z})=\lambda [\tilde {z}\sqrt {\tilde {z}^2+1}+\ln (\tilde {z}+\sqrt {\tilde {z}^2+1})]+{\rm \pi} /4$. Thus at the interface, we have
where
$Y_{0}={\rm \pi} /4$.
After some trivial calculations, we find that the transmission ratio at the interface is always
$\eta \equiv 1$.

























































